WhatsTheBigData

[April 22, 2019]

“The past fifty years,” says Dr. Eric Topol in Deep Medicine: How Artificial Intelligence Can Make Medicine Human Again, “have introduced important changes to radiology. As the medium moved from analog to digital… the whole process, extending from films to CT, PET, nuclear, and MRI scans, has been made more efficient. Except the interpretation.”

Dr. Topol quotes studies suggesting that errors in interpretation of medical scans “are far worse than generally accepted,” with false positive rates of 2% and false negative rates over 25%. As a result, 31% of American radiologists have experienced a malpractice claim, “most of which were related to missed diagnoses.”

The rapid advances in computer vision due to the application of AI starting in 2012, have led to predictions of the imminent demise of radiologists, to be replaced by better diagnosticians—deep learning algorithms. Geoffrey Hinton, one of this year’s Turing Award winners and a major contributor to the remarkable success of deep learning, suggested in 2016 that “People should stop training radiologists now. It’s just completely obvious that in five years deep learning is going to do better than radiologists.” In the same year, an article published in the Journal of the American College of Radiology warned that “The ultimate threat to radiology is machine learning. Machine learning will become a powerful force in radiology in the next 5 to 10 years and could end radiology as a thriving specialty.”

While Dr. Topol believes that eventually all medical scans will be read by machines, he argues that radiologists can have a bright future if they “adapt and embrace a partnership with machines.” Eyal Gura, co-founder and CEO of Zebra Medical Vision, agrees: “AI can help doctors get to the right place quickly and make the right decision.”

Gura’s vision is that Zebra will help “automate every visual aspect of medicine,” going beyond radiology to pathology, dermatology, dentistry, and to all situations where “a doctor or a nurse are staring at an image and need to make a quick decision.” This “automation” does not mean replacing doctors. Rather, it means the augmentation of their work, providing consistent, accurate, and timely assistance. “We need all the doctors we have in the world and we will need 10X more because of the aging population,” says Gura.

Zebra’s work experience with radiologists in more than 50 hospitals worldwide highlights the role of AI as Augmented Intelligence. Its algorithms help overcome the “training bias,” the fact that “their brains are fine-tuned” to the specific cases they studied in their textbooks, says Gura. “Once trained,” writes Dr. Topol, “doctors are pretty much wedged into their level of diagnostic performance throughout their career. Surprisingly, there is no system in place for doctors to get feedback on their diagnostic skills during their careers, either.”

Zebra’s algorithms provide this missing feedback by offering radiologists a second opinion. In addition, they provide assistance and augmentation when there’s minimal or non-existent training. Consider the case of a young doctor in the ER who is not familiar with how a tiny brain-bleed looks on a scan and entirely misses it or a rural clinic with no access to a radiologist (the World Health Organization estimates that two-thirds of the world’s population has no access to any diagnostic imaging).

So far, Zebra has developed 48 algorithms addressing 48 different medical conditions (8 have already received regulatory approval in Europe and one in the US) that assist radiologists at different points in time, from acute conditions to current diseases to preventive medicine based on past scans. Earlier this year, Zebra announced the first multi-modality AI triage solution, addressing two life-threatening conditions, brain-bleeds and pneumothorax (the presence of gas between the lung and the chest wall). The Zebra triage solution is integrated into the hospital’s workflow, sends an alert when it detects a suspected acute finding, reducing the time to diagnosis by 80%.

At the other end of the timescale, Zebra’s algorithms can help in reviewing past scans, identifying patients at-risk and assisting in population health management.  A number of 5-year retrospective cohort studies conducted by one of Zebra’s research partners, Clalit Research Institute, found that Zebra’s algorithms performed better than the current medical gold standards for predicting osteoporosis fractures and risk for cardiac event.

Last month, Zebra announced it will collaborate with HealthNet Global (HNG), part of the Apollo Hospitals Group in India, to “provide timely, cost-effective, quality care to patients in remote and rural locations.” For example, they plan to develop a chest X-ray interpretation tool for TB to help in its early diagnosis by supplementing sputum testing which is only 50% accurate and frequently misses the disease in its early stages (the World Health Organization estimates that 3.6 million people with TB are missed by health systems every year and do not receive adequate care). HNG and Zebra will be supported by a grant from India-Israel Industrial R&D and Technological Innovation Fund.

“In rural areas in India you will be able to have a nurse and an X-ray technician and get an early diagnosis or an alert on an acute condition to allow them to provide the first line of support,” says Gura. Thanks to Modicare, “out of nowhere, 500 million people in India will have health insurance, but you will not have more doctors to treat them,” he adds, promising a similar deployment by Zebra in Africa later this year.

Headquartered in Israel, Zebra most recently raised a $30 million Series C in July 2018, led by aMoon Ventures, with participation from Aurum, Johnson & Johnson Innovation—JJDC Inc., Intermountain Health (also acting as one of Zebra’s data and research partners ) and AI pioneers Fei-Fei Li and Richard Socher. Existing investors Khosla Ventures, Nvidia, Marc Benioff, OurCrowd and Dolby Ventures also returned for the round, helping bring total funds raised by Zebra to $50 million.

The funds will be used to further improve Zebra’s algorithm development process, commercialization of these algorithms (designing and launching products), and integration with health providers’ existing systems, all crucial to achieving Zebra’s goal of becoming a one-stop shop (at $1 per scan), and establishing a sustainable competitive advantage.

Zebra is part of a growing community of Israeli digital health companies. Last year, total investments in the sector increased 32% and exceeded $500 million for the first time, with 85% of this amount going to companies utilizing AI solutions, according to Start-Up Nation Central. A significant competitive edge for these startups is the availability of data collected over the last 25 years by Israel’s four HMOs and their affiliated hospitals, serving 98% of the population and using the same electronic medical records system.

Israeli health-related startups (more than 1,200 in digital health, medical devices, and Pharma), with their unique mission and potential for making a real difference are increasingly attractive to Israeli AI, machine learning, and data science experts, now being assiduously courted by deep-pocketed global competitors. “Especially at a certain age, they feel the need to do something more meaningful. They see that the time and talent they spend on ad conversion can be better spent on saving their mother or father,” says Gura.

Originally published on Forbes.com